Gaining Advantage in a Forensic Match GameJanuary 2016
Topics: Modeling and Simulation, Image Processing, Biometrics
TV crime dramas make it look so easy: A sharp-eyed detective lifts a partial fingerprint off a water glass or a book jacket or a door latch. She sends the sample to a lab where, with a few keystrokes—and before the next commercial—the investigators are slapping cuffs on the suspect.
In the real world, matching latent, or unintentionally deposited, fingerprints to the human who left them is a lot more complicated. Identifying the owner requires matching the evidence—often smudged, incomplete, or deposited on top of other markings—with complete prints on file in one of a number of Automated Fingerprint Identification System (AFIS) databases.
A student team at Harvey Mudd College (HMC) in Claremont, Calif., has developed a scoring system for latent fingerprints that can make the matching process more accurate and efficient. And MITRE is helping them.
Sharpening the Focus on Identity Verification
The students are part of HMC's Clinic Program, which challenges them to solve real-world problems using computer science and math. The students choose their research topics from a list of problems submitted by corporations and research centers, including Aerospace, Nike, Disney Animation Studios, SpaceX, and Lockheed Martin.
The HMC Clinic reflects the educational philosophy held by its founder, Jack Alford, HMC engineering professor emeritus. Alford likened engineering education to dancing: "You don’t learn it in a darkened lecture hall watching slides. You learn it by getting out on the dance floor and having your toes stepped on."
MITRE's liaison to the program is Nick Orlans, a biometrics specialist. He has helped formulate research strategy at MITRE and also supported various challenge problems for identity recognition and intelligence. Two years ago, a Clinic team developed a flexible, simple, efficient framework for evaluating tattoo-matching using computer-vision algorithms. In the 2015-16 academic year a Clinic team is exploring ways to achieve identity recognition when the subject is wearing make-up to change his appearance.
This year, Orlans brought a problem involving partial fingerprints. 'Latents are high-value artifacts," he says, 'but they tend to fall into two categories: either high quality and well-suited for automation, or lower qualities such that resources of human experience and expertise are needed to obtain some degree of partial automation. There's never been a good method for assessing when prints will be suitable for full automation, or fall somewhere in the realm between semi-automated and usable. Knowing which latent prints will perform equally well in all systems or if some are best suited for a particular industry solution is also a challenge."
A Collaborative Approach to Big Data Analytics
Every student at HMC has to participate in at least one Clinic challenge. They work with students from other majors to bring a multidisciplinary approach to the problem.
The MITRE project required students to evaluate existing analysis processes and develop a mathematical model for scoring latent fingerprint quality. They also had to identify a way of validating whether the quality metrics could predict "matchability." The group briefed a MITRE team on its findings at the end of the school year and published a white paper to back up the work.
The process typically involves examining the print at increasing levels of detail.
- The first step detects general patterns, like loops, whorls, and arches, as well as the direction a print's ridges face.
- The second phase looks at minutiae: where the ridges begin, end, or split. Initial analysis and comparisons takes place at this level.
- A third phase can identify features as minuscule as ridge contours and sweat pores, which are observable only in high-resolution images. Few latents are pristine enough to hold up to that level of examination.
The HMC team's scoring system rejects the lowest-quality matches. Those with high scores are better candidates for comparison with AFIS databases. Applying the algorithm allows analysts to begin their work with fewer, but better quality, latents that have a better chance of identification.
Finding the Perfect Match
Ultimately, success of any identification effort depends on the latent matching an exemplar print on file in an AFIS database. The federal government's AFIS alone holds over 100 million fingerprint records. Law enforcement retains prints on criminal suspects or convicts, and military systems store prints from detainees and humanitarian efforts. Other systems include prints provided by people entering the United States, entering the military, holding certain government or security jobs, or other situations that require identification.
"It's amazing when you think that before the 1990s, fingerprints were on cards in file cabinets," says Margaret Lepley, a MITRE principal systems engineer who specializes in image processing and fingerprint imaging-and-capture devices. "Examiners had their ways of coding and filing so they knew where to start searching for a match, but it required a lot of manual work. Now computers are doing some of that work, but humans still step in to assist the initial markup, and evaluate the computer-generated list of candidates."
Digitization has helped a lot. And the students' work could help even more.
Government and law enforcement have files containing millions of unidentified latents. Sarah Doyle, a senior biometric engineer at MITRE, said the students' scoring system shows promise for eliminating those with little chance of yielding matches. Eliminating them would reduce the number of latents awaiting manual review.
Orlans said MITRE’s participation in the Clinic helps strengthen the company's reach into academia, too. The fresh approaches and new ideas that come from university partnerships strengthens the depth and capacity of work MITRE can take on.
It can also help attract top young talent to the company. "We're very open about that," Orlans says. “We want them to know about technically challenging problems that need better solutions."
—by Molly Manchenton
Are you a good fit for MITRE, too? Explore our current Job Openings.